Fig. 6
From: Dataset-centric evaluation of federated intrusion detection models in IoT networks
Normalized confusion matrices for the TII-SSRC-23 test set. The five-class matrices illustrate the greater difficulty of this dataset. Per-class confusion explains the modest gap to Edge and CIC reported in Table 7. Mis-classifications mainly occur between InfoGather and DoS, and between Mirai and DoS. FedProx-Transformer yields the sharpest diagonal, evidencing its robustness on heterogeneous data.
